HOMOMORPHISMS OF FILTERED MODULES BY ALEX HELLER 1. Introduction. The following simple problem appears to be without a known solution. Let A', A", B', B" be modules and suppose a eExt'04", A'), ß e ExV(B",B'). Then a and ß determine extensions A of A" by A' and B of B" by B'. How may the group Hom(/l,B) be expressed in terms of the four modules and two extension classes? We give here a partial solution of this problem. Hom(A,B) is seen to have a canonical filtration; we compute the associated graded group. Our failure to compute Hom(y4,B) itself is to be thought of as analogous to the failure of the Kiinneth theorem to compute canonically the homology of a product of chain complexes (cf. [2]). This partial solution is a special case of a solution of a more general problem. If A and B are filtered modules then how may tlom(A,B), or rather, its associated graded group, be computed in terms of the associated graded modules A" and B" of A and B, together with the "extension classes" needed to reconstruct A and B from these graded modules? Indeed a similar question arises with respect to all the groups Extq(A,B). We shall see that associated with such data there is a spectral sequence whose term E1 is just Ext(A",B"), and whose term E °° is associated with a filtration of Ext(A,B). The derivation d1 is easily computed. No attempt is made to com- pute the higher derivations in the general case. However, in the special case with which we began, d2, which is the only other nonvanishing one, is explicitly computed as a secondary operation on Ext(^l",B"). Our attention is restricted here to modules with finite nitrations. It is not par- ticularly difficult to generalize the results to arbitrary increasing nitrations of modules. However the results stated here go over without any essential change to nitrations in arbitrary exact categories, or indeed in additive categories with abelian structures [1]. This would not be true of infinite nitrations. The restrict- ion is adopted for this reason. It is proposed to use the information obtained here in the study of universal coefficient and Kiinneth theorems for chain complexes of modules. The present study is however completely independent of such application. Since it throws some light on the problem originally mentioned it seems appropriate to present it separately. Received by the editors April 20, 1962. 49 License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use 50 ALEX HELLER [April 2. The category of filtered modules. We shall for the whole of this paper suppose that there is, somewhere in the background, a ring, that all modules considered are left modules over this ring, and that all homomorphisms of mo- dules are operator homomorphisms with respect to this ring. By a graded module we shall mean a sequence of modules (indexed by nega- tive as well as positive integers); we refrain from identifying it with the sum of its terms. For filtered modules we adopt the convenient if somewhat uncon- ventional definition of [3]; a filtered module X, or filtration X is a short exact sequence X = (0 -+ X i X î» X" -* 0) of graded modules, with x, x" homogeneous of degrees 1,0. The graded module X" is of course the associated graded module of X. The filtration is increasing if X„ = 0 for some m (and thus all k ^m) and decreasing if for all k greater than some n the modules X"k are 0. If it is both it is &finite filtration. In this paper we restrict ourselves to finite filtrations. For such a filtration X (and indeed for any decreasing filtrations) we have, for sufficiently large k, Xk x Xk+J » •••; we denote the common value by Xx and say that X is &fil- tration of Xœ. If X and Y = (0 -> Y^ Y% Y" -►0) are filtrations we define the graded abelian group Hom(X,Y) by letting Homg(X,Y) be the set of pairs (/,/") of maps, each homogeneous of degree q, such that commutes. With this convention the filtered modules clearly become the objects of a graded additive category J5". This category is never exact, but it has a natural abelian structure (cf. [1]) given by the convention that (/,/") is a proper mono- morphism (epimorphism) whenever both / and /" are monomorphisms (epi- morphisms). We shall always consider J5" with this structure, which is perhaps more neatly characterized by the following lemma: Lemma 2.1. (/,/") :X-> Y is a proper monomorphism (epimorphism, iso- morphism) if and only if f" :X"-* Y" is a monomorphism (epimorphism, isomorphism). This follows immediately from the finiteness of the filtrations. We denote by S" the functor S"X = X", S"(f,f") =/", which maps & onto the category of finitely graded modules. It is, by 2.1, exact. If X" is a finitely graded module an extension of X" is a filtration X e & with S"X = X". If X and Y are extensions of X" we say that they are equivalent if there is an f:X-+ Y with S"f = 1 : X". The set of equivalence classes of such extensions is denoted by QX . If X is such an extension it determines an extension-class eXeSX". License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use 1963] HOMOMORPHISMS OF FILTERED MODULES 51 We shall also want to use the above notions for graded modules. A filtered graded module X is a short exact sequence 0-»!^X*-> X" -* 0 of bigraded modules with maps x of bidegree (1,0) and x" of bidegree (0,0); we refer to the two degrees as filtering and auxiliary degrees. The finiteness conditions are of course confined to the filtering degree. We content ourselves with asserting them in the strongest possible form: for some m, all Xim = 0, and for some n and all k> n, all X"k = 0. More generally, we may even discuss filtered multigraded modules, subject to the same conditions, which we describe by saying that the finiteness condition holds with respect to the filtering degree uniformly in the other degrees. 3. Splittings, projectives and the filtration of Hom(X, Y). If X is a filtered module, a left-splitting of X is a map £ :X -> X homogeneous of degree —1 such that <j;x= 1 : X. The set of these (which is often empty) we denote by *PX. A right-splitting of X is a map {" : X" -> X, of degree 0, with x"£" = 1 : X"; the set of these is *P*X. A bijective correspondence <jj-»£,* of *PX with f/*X is given by the condition ¿¡¿j*= 0, which implies that (x, £,<!;*,x") is a direct sum decom- position of X as a sum of X and X", and in particular that x£ + £*x" = 1 :X. Lemma 3.1. IfÇeVX then 1 :X = £,•„ *"€**"$"■ Observe first that the sum on the right is in fact finite on each Xq. Now re- writing £*x" as 1 : X — x£ we have OO OO OO I xni*xT = I x"f - I xT 11=0 »=0 »= 1 which is of course sufficient. Corollary 3.2. // also neVY and (/,/") :X-> Y then f = Z t ynn*yYfxkt*x"e. n=0 k=0 We need only observe, in order to see this, that terms with k > n vanish. It is worth remarking that /'/£* =/" so that the term with n = k = 0 is y*f"x". Now for any X,Y e !F we have an exact sequence (3.3) 0 -» Hom(X,Y)h Hom(X,Y)X Hom(r',y") given by h(f,f") = (yf,0) = (/x,0) and h(f,f") =/". Here h is homogeneous of degree 1 and h of degree 0. Thus Hom(X,Y) is in a natural way a filtered group or, more precisely, there is a filtered group (3.3') 0 -+ Hom(X,Y)A Hom(X,Y)t Hom"(X,Y)-* 0 whose associated graded group Hom"(X Y) is a subgroup of Hom(X",Y"). The finiteness conditions clearly imply that Homœ(X,Y) « Rom(X00,Y00), so that (3.3') is a filtration of Hom(Xao,Yx). One of our principal aims here is the com- putation of Hom"(X,Y). License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use 52 ALEX HELLER [April The first step is to determine the projectives of #\ Proposition 3.4. i/X is a filtered module then the following are equivalent: (1) X" is projective; (2) the map h of 3.3 is, for all Y, an epimorphism; (3) X is projective. We shall prove (1) =>(2);(2)=>(1); (1,2) =>(3); (3) =>(1). Suppose first that X" is projective. Thus X splits; let £,e^X. If/" :X"-> Y" there is a £:X"^ Y with y"g=f. Set /: £,"<>/£*"£". Then (/,/") :X->Y. Now suppose X has the property (2). Define X = (0 -> X-> X -> X" -> 0) by -S",= Z* = -x^b, the maps in X being injections and projections. Then there is a map (/, 1 :X"):X-+X which by 2.1 is an isomorphism. Thus X splits; let ^eTX. For any q let 0-» y,_i-» Y4->-Xs"->0 be exact and define a filtered module Y by setting Yr = 0, r < q - 1 and Yr = Y?,r > #. The associated graded module of Y is Xq in degree q, Yq_L in degree q - 1 and 0 elsewhere.